aiqbench analysis
This article is primarily about AI: it proposes a training method that aligns multimodal LLMs with human similarity judgements, using a behavior-derived embedding space and dimension-grounded rationales. That makes it an AI research paper, so the topic gate passes. The results are modest and benchmark-scoped: consistency with human judgements improves while general multimodal performance stays competitive, and the fMRI RSA evidence is correlational, so claims about shared cognitive structure should be read cautiously. Readers should watch whether the gains hold on independent tasks and whether the rationale-training approach generalizes beyond the odd-one-out setup.
Aligning MLLMs with the Latent Structure of Human Cognition via Behavior-Derived Semantic Dimensions
arXiv:2610.41951v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) have achieved impressive performance across a wide range of reasoning tasks, yet it remains unclear whether their internal representations and decision criteria align with the latent structure of human cognition. In this work, we propose a novel framework to align MLLMs with the multidimensional mental representations underlying human similarity judgements. We build on a previously established behavior-derived semantic embedding space, in which objects are represented along 66 interpretable dimensions derived from 4.70 million human Odd-One-Out (O1O) judgements. For each triplet, we infer the most salient decision dimension and convert it into a dimension-grounded linguistic rationale. This enables MLLMs to learn both the final choice and a justification consistent with the inferred behavioral criterion. Experimental results demonstrate that our approach improves model consistency with human similarity judgements while retaining competitive performance on standard multimodal benchmarks. Furthermore, using searchlight Representational Similarity Analysis (RSA) on an independent fMRI Natural Scenes Dataset (NSD), we observe increased representational alignment between model and human. Ultimately, our findings point to a data-driven route for incorporating human cognitive structure into MLLMs alignment. The code is available \href{https://github.com/ening-open/cogalign}{here}.
Abstract:Multimodal Large Language Models (MLLMs) have achieved impressive performance across a wide range of reasoning tasks, yet it remains unclear whether their internal representations and decision criteria align with the latent structure of human cognition. In this work, we propose a novel framework to align MLLMs with the multidimensional mental representations underlying human similarity judgements. We build on a previously established behavior-derived semantic embedding space, in which objects are represented along 66 interpretable dimensions derived from 4.70 million human Odd-One-Out (O1O) judgements. For each triplet, we infer the most salient decision dimension and convert it into a dimension-grounded linguistic rationale. This enables MLLMs to learn both the final choice and a justification consistent with the inferred behavioral criterion. Experimental results demonstrate that our approach improves model consistency with human similarity judgements while retaining competitive performance on standard multimodal benchmarks. Furthermore, using searchlight Representational Similarity Analysis (RSA) on an independent fMRI Natural Scenes Dataset (NSD), we observe increased representational alignment between model and human. Ultimately, our findings point to a data-driven route for incorporating human cognitive structure into MLLMs alignment. The code is available \href{this https URL}{here}.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2610.41951 [cs.CL]
(or arXiv:2610.41951v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.41951 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 284 words